• Iman Nabiyouni

Power line Maintenance - A Machine Learning Use Case


Electricity plays a vital role in our daily lives. We expect continuous and uninterrupted power, but we often overlook the behind-the-scenes work that goes into keeping the lights on and the grid running. One of the most critical aspects of this work is the maintenance required for the physical generation, transmission, and distribution of energy infrastructure. This maintenance is not only needed to reduce the likelihood of outages and network failures, but also reduces the likelihood of grid-initiated wildfires such as we have seen raging across the West Coast in the past few years.



Maintenance of Electrical Power Lines

A key component of conducting optimal maintenance is early detection of faults along energy infrastructure components as a part of the inspection process. This process is currently done manually with hundreds of linemen and field technicians sifting, tagging, and labeling thousands of visually captured images from drones, helicopters, and aircrafts. These inspections are costly and time intensive. They are prone to human errors due to the vast amount of data that must be manually tagged.


Artificial Intelligence to the Rescue!

Artificial Intelligence (AI) has opened a new world of possibilities for analyzing inspection data faster, cheaper, and more efficiently than ever before. Now, linemen and field technicians can optimally plan and conduct maintenance after inspection processing.


What is Artificial Intelligence?

AI is commonly recognized as a form of computer science that is designed to perform tasks which traditionally required human intelligence. It can create a scalable process at a level far more efficient than humans. This replication of intelligence from machines makes life easier for humans by providing similar services that are remarkably faster and lower in cost.


How Does It Work?

In order to understand how machine learning works, it is important to learn the process of AI for task automation. With each task where there is desired automation, the first step is relevant data collection and transferring, with the data consisting of things such as images, audio, text, etc. This data is digitized by converting it into a number format which is readable by the computer. A computer model is applied to the data to produce some results (i.e. predictions). These models can be considered “black boxes”, which have an input and output. Specifically, these models are collections of variables (weights) that are applied to the digitized input data. An interpretable example (of these models) is matrices which are operated on another matrix (digitized input) and produce some numerical values (output). These numbers will be converted into human readable values as a result of the model prediction (figure 1).


Figure 1. Artificial Intelligence model as a black box which operates on a digitized image for producing & predicting some outputs


What Is New About Machine Learning?

Machine learning (ML) is considered a form of AI which has become quite popular in the past two decades. The main difference between AI and ML is the trainability of these models. ML models are trained and improved over time in contrast to AI models, which are manually created. One example of AI models is called Expert Systems, which is used for medical purposes such as creating Antibiotic prescriptions. This prescriptive model is usually presented as a decision tree. In each node, the model asks questions about the condition of the patient and recommends a prescription for the patient (figure 2).


Figure 2. A manually created Artificial Intelligence model for prescribing Antibiotics


One of the successful branches of ML is Deep Learning, which has become popular in recent years and mimics human brain processes. In this context, a neuron is defined as a decision maker which receives an input signal and produces a response based on that input. A connected network of these neurons is called a deep neural network (DNN). The process of training and working with these algorithms is known as deep learning.


How to Process Images with Machine Learning?

One use case of deep learning and neural networks can be found in Computer Vision - which is the processing and high-level understanding of images. Most of the successful computer vision structures are based on deep learning and neural networks.


What is Image Classification?

The main subfield of computer vision is Image Classification in which image-based content is classified to a predefined class. One of the most popular approaches in this area is employment of a deep learning algorithm called Convolutional Neural Network (CNN). A very famous application of this technique is using automated handwriting analyzer in postal services (figure 3). The model that is used in these analyzers is usually a combination of multiple neural network layers. In some of these layers, a convolution takes place which is the application of a rectangular filter to an input.

Figure 3. Character recognition with convolutional neural network (CNN)



What is Object Detection?

The next important subfield of Computer Vision is object detection. In this approach (similar to the other types of machine learning), some learning algorithms (models) are trained over a considerable number of labeled images which are analyzed by the models and provide predictions based on the images. These predictions include location and type of visual objects that are detected in the image. In each image, the predictions are visualized by drawing bounding boxes around the detected objects (figure 4).

Figure 4. Extracting position and labels of the object which is visualize with a bounding box in Object Detection



How Buzz Solutions Inspects Power Line Images with Computer Vision?

Object Detection is usually used for automation of businesses worldwide. It is well-known mostly for use cases such as self-driving cars (Tesla), face detection (iPhone), and image auto-tagging (Facebook). Application of Object Detection is increasing in many industries. An example being in retail and E-commerce with recent trends using ML for warehouse automation (with AI-powered robots). In banking, AI is used for scanning documents such as deposit checks and authenticating with biometrics such as face or voice authentication. In Healthcare, it is used for processing MRI images, genetics analysis, and pose estimations for fitness and wellness.

This approach can also be used for the inspection of power lines. This is where Buzz Solutions plugs in to automate the analytical portion of power line visual inspections. This is done by processing the images captured from electrical facilities. Our algorithms are trained leveraging historical datasets with labeled faults. With our highly accurate and continuously learning algorithms, we can create real value through the use of AI to identify faults and failures within the critical energy infrastructure. These labels in the power sector include objects such as rust, broken infrastructure components, and vegetation surrounding the line - just to name a few. An example of such predictions is shown in the image below.



That being said, this is only the beginning of possibilities in leveraging AI in the power sector. Not only can this solution drastically reduce costs and time at scale, but our algorithms continue to improve in accuracy over time becoming more efficient, accurate, and highly personalized to each utility dataset. Utilities can use these insights to track asset health over time, enabling them to better understand the highest risk areas, rate of failures in components, and more accurately conduct maintenance. With AI, utility engineers and inspectors can make decisions faster and utilize inspection data to come to more informed decisions.


In conclusion, applications of machine learning are growing fast as it facilitates the tasks and makes them more affordable for small businesses. In particular, object detection - which is the most popular application currently in computer vision, has successful use-cases, and will help the automated machines have a better understanding of their visual targets/environment.



- Iman Nabiyouni

Senior Machine Learning/Computer Vision Engineer

Buzz Solutions, Inc.


 

Enjoyed reading this article? Share this with your colleagues.